Google has reduced the cost of AI question answering by 80%!

The AI battle among tech giants has entered a fierce stage, with Google deploying cost reduction and efficiency enhancement strategies.

Last week, after Google's parent company Alphabet released impressive financial results, its stock price surged by 10% at one point. What excited investors was not only Google's first-ever dividend payout and increased stock buybacks, but more importantly, Alphabet's CEO Sundar Pichai revealed that Google successfully reduced the cost of AI chatbot Q&A by 80% in the past year!

Pichai's statement may be more significant than any other content in Google's financial report.

It's worth noting that Google's search engine business contributes half of the company's revenue and is a major source of its profits. However, when Microsoft added AI chatbot and webpage summarization features to its Bing search engine in early 2023, investors worried that Google's search engine business might be eroded by Microsoft. Moreover, the continuous increase in Google's AI investments could lead to a decline in overall profits. Google's board chairman John Hennessy warned last year that the cost of using AI chatbots for Q&A could be ten times that of traditional Google search.

Now, during last week's earnings conference call, Pichai stated that Google has effectively reduced Q&A costs through "technological breakthroughs" with the Gemini AI model and custom AI chips.

"People question whether the cost of AI search Q&A services will be high, and we are very, very confident that we can control this cost."

His remarks reassured investors, and with Google's search market share remaining almost unchanged over the past year, it indicates that these concerns may be unnecessary.

Google is serious about cost reduction and efficiency enhancement in AI chatbots

In order to lead in AI chatbot technology, Google has been making strategic moves for a long time.

Last May, Google began rolling out AI-generated search results to some test users. Unlike previous tests that presented users with a bunch of web links, this test generated answers based on webpage information. (Users can still scroll down the page to find links or click on the information sources referenced in the AI answers.)

By March this year, Google had started incorporating AI answers into its search function. However, due to the high cost of running AI models, Google initially stated that only a small number of users would have access to this feature.

Despite Google making only "small" progress in AI search Q&A, a lot of preparation has been done behind the scenes.

After prioritizing AI as a strategic goal, Google reorganized its search engine department in March this year, bringing in a large number of senior professionals in the AI field. Liz Reid, responsible for developing new generative AI features, was appointed to lead different search teams, while Cheenu Venkatachary, with expertise in large language models, was promoted to head search quality and ranking operations. Google hopes that the search department and the AI research team DeepMind can collaborate fully to make AI Q&A search results more accurate Furthermore, Google has been under pressure to reduce costs and launch advanced AI products over the past year. Last week, Google stated that compared to a year ago, the company's capital expenditure has doubled to $12 billion, mainly due to investments in AI technology infrastructure.

An individual involved in Google search work mentioned that to reduce costs and increase efficiency in AI search, they use different levels of AI models to answer different types of queries.

For relatively simple queries, a lower-level Gemini version is used, requiring less computing power. More complex queries, on the other hand, will require the use of the higher-cost flagship model, Gemini Ultra.

If Google decides to fully integrate AI capabilities into traditional search engines, it could potentially save billions of dollars overall.

However, even though the cost of AI question-answering has decreased by 80%, the cost of using AI for search is still higher than regular search.

Therefore, how to make money with the expensive AI that has been developed is something Google still needs to consider.

How to make money with AI?

In this regard, Google aims to organically integrate advertisements into AI-generated answer results, mimicking its traditional search engine profit model. However, the results seem to be unsatisfactory.

According to two individuals who have worked at Google, because AI summarizes answers more concisely than traditional search results (which link to articles that may provide answers), people searching for information may not need to click into another page to get the answer, affecting the effectiveness of ads displayed based on search engine traffic dependence.

A year ago, Google experimented with displaying ads above AI question-answer results. For example, when inputting "how to buy a bracelet" into AI, a group of ads appeared at the top, followed by the AI-generated answer below. However, the effect seemed inconclusive.

Pichai pointed out that at least one positive sign is that people are now using AI search more frequently, which is a good sign for Google's search advertising business. More user searches mean more opportunities to display ads